Ok well here's the short and long if it:
In short, yes the 686B works, and unlike the VT6421A, it passes TRIM. And for the long of it:
Well, first of all, Windows 98 proved once again that a preliminary Prime95 run under a Linux floppy micro-distro is no substitute for a proper stability test... My mobile 2400+ is not in fact stable at 2133 at 1.55 volts. I just lowered the FSB to 100Mhz instead of experimenting with it at this time, so I could get on with my testing. I may purchase a mobile 2500+ in the near future.
Now, while the 686B does work fine with the aforementioned combo, it does not do so without a caveat or shall we say altogether well. These screenshots speak for themselves (sorry that they're GIFs, I was limited to the primitive conversion filters in 9x MS Paint):
Marvell controller with ICH2:
ICH2_MARVELL2.gif
Marvell controller with 686B:
686B_MARVELL_2.gif
So the 686B has only half of the write performance to that of the ICH2 in this scenario.
Still, I would personally rather use the 686B than an add-in Promise Ultra133TX2, simply because it is generally always better to use the native implementation than it is to use an add-in card, and the speed is still quite acceptable (and not to mention, the read speed is perfectly fine, only the write speed is affected).
VIA 4.43 was installed, and DMA was enabled.
But I'm going with the P4 system over the KT7A, simply because ODCM throttling is so good, the KT7A simply cannot compare with its flexibility in that regard. And my P4 board has SB-Link, so I'm covered there.