IMHO, the concept of adding "speed bumps" to the game was a mistake from the start. The next time, you might want to give more consideration to your new players (customers) who are already so shocked at the game's complexity that they would welcome an easier military aspect, instead of the "generals" who keep insisting that you make the military game harder.
Well there's actually a few factors that affect these points - one is the fact that Garrisons were meant to replace the building and positioning of conscript infantry units by 'micro management'. Both the AI and many human players would use this to create those very same 'speed bumps', so why not make the process easier; this should in a way reduce complexity by automating the process. Plus, Garrisons that are dormant are a zero CPU and memory drain, since the only wake up and act as units when enemies approach the city, and as such are much more efficient than manually placed conscript units.
As well, the AI use of Garrisons is directly affected by the military difficulty levels, and there is much less use of them at the easier levels for new players.
But the garrison problem is compounded by other questionable elements in the game:
1. Over-population of units.
It is true that this is a problem still, though we are working on it. The sale of older units between regions actually helped this a bit, sending unused units into war zones, but there's more to do.
2. Short-range units allowed to attack outside of their hex.
The tactical design of SR2020 is that units generally fight against adjacent hexes, only in rare conditions do they fight 'in hex'. As this is the design, it means that units are generally expected to be 'near the edge' with each other as they battle. This is also backed up by the 'standoff' unit policy, which is that artillery and such can remain untouched as long as an offensive unit like infantry or a tank are in the hex. It's not perfect, but it works pretty good.
3. Close combat penelties.
Some units need their stats better balanced for this, and the AI needs to understand close combat better, but again I think it is an engine concept that generally works.
4. No-miss attacks.
As has been discussed elsewhere, a 'random' factor is less necessary when the combat design is as detailed as is the case in SR2020; the damage from attacks builds up over dozens (or hundreds) of engagements in each day.
5. Overly-strong garrison units.
Yes, I agree here, the Garrison stats need adjustment.