Well then, into the breach.
1. This is a good effort to standardize the rules, but there is a reason robots had a higher action check score. It's the processor, and no further explanation is required. Brain is slow, and robot that has -d6 action check modifier should of course gain those 4 actions per round. I say leave the dataware numbers, they do (in this instance) make sense.
As for gameplay, some might argue 4 actions per round (or 3, or 2, depending on PL and dex of course) is overpowering. I strongly dissagree. For example a human with the equivalent int (PL8 amazing proc) would gain (by variant rules of course) 83 skill points to distribute, while in any normal case (if you're building a robot hero, i' not talking of general usage droids here) a robot after he buys legs, hands, processor, and other peripheralia would have 40ish points to spend, and is still highly handicapped by the memory space etc. But this would be better left for another topic.
2.I was speaking of the non-AI robot perks/flaws. From fuzzy logic to aasimov's circuitry, those perks and flaws are the very flavour of the robot hero.
SQuAT advice: Put in all the standard dataware ones, and leave space for customization (as always, XML files are there). Also have in mind to recalculate available memory slots: -1 for OS (always), -1 extra for redundant circuitry (for example) etc.
There are some major mistakes in the dataware which concern some flaws and perks but those can be addressed later. For example, the emancipated is woefully underexplained, and contradicts previous text (4 vs 5 SP), photographic memory (listed as available for robots) makes no sense and should be removed (especially when the robot has life recorder), etc etc.
3. This way of thinking has one major flaw. By specifically distributing attributes, like that dex 8 for example, you make a major problem when creating a robot hero. True, you ease much on designing standard battle droids etc, but it has no connection any more to the 50 attribute points to be distributed across 5 areas. Now, why is this bad you may ask. It's bad because when a player designs a robot he's either underpowered or overpowered compared to non-robot partymembers.
Dataware addressed this problem with a dual self-control system: 1. there were defaults a robot had and 2. each upgrade cost either SP or money. This system has merit, and should not be rejected, only modified.
First I need to further elaborate on the *reasons* behind dataware system:
By spending SP (or money) a player picks ranges for the attributes, not the attribute itself. Even chasis should have a range since not all frames are as hardy as other. Leave it up to the player distribute attribute points, otherwise all robots (heroes, i stress again) of same height (for instance) and PL would be essentially identical, as all would have the best proc for the PL and thus have same con, int, and most probbably would choose same actuators and thus have same str and dex. This is very very bad.
To fix the problem I suggest that dataware's system be left, perhaps not with the same ranges but most certainly with the same idea. To sum it up - leave dataware, and fix the ranges. A player will be able to upgrade the droid through standard upgrades, which you already envisioned, but those upgrades should be very costly .
Or, alternativly, make defaults for certain progress level which sum at 50 attribute points as before. At character creation have a player pick upgrades (up to the max allowed for a certain robot part), but if he chooses to upgrade a stat it should be as costly as before - at perk value, 10SP or more per atribute point.
4. There are a few methods available to determine what should be done with these.
a) Have the proccessor upgrades be logical, rather than physical. They would be available for per and will only. They would eat lots of memory space (2per=4mem slots). They would essentially be programs, which money can buy. Don't sell them for SP. This is a method I'd use if you stick to dataware creation system, which I do recommend. This way you also solved the limit problem, since a player with +3 per and +3 will, using a PL8 amazing proc has mere 5 available slots for skills. (18 available -1 OS, -12 upgrades = 5)
Have the chasis upgrades be physical, as additional actuators to increase strength or dex. These should also cost a lot of money. The limit problem could be solved that they count against action check and action check modifier (if talking of heavy strength increasing actuators) , or they can count against maximum carry weight and frame studriness (if talking of dex increasers which eat up frame space, making it ligher and less hardy). For example +2dex = -1con + carry weight modifier is reduced by 2.
c) don't make them eat memory space and other things mentioned above, rather count them against perk/flaw count. This is basically a robot version of heightened ability perk non-robots have.
4. A person who wants to firewall a robot certainly won't install a nijack. He'll simply remove the wireless connector and that's it. Or, more simpy, install a manual on/off switch on it. When he wants to connect to a gridcaster he turns it on, and goes online, and problem solved. Or alternatively, uses a socket connector. It's there for a reason.
I'll let you think about these issues for a while.