origin of the problem:
in this previous thread, several people told me that the ibm vga and some svga cards, including the trident8800, can only process 4bits of video data per cycle before sending it to ramdac, and discussed it. this issue causes 256color modes to require double the clock frequency compared to the same resolution in 16color modes.
regarding ram size, 256kb is expected to support 640*400*256color mode, while 512kb is sufficient for 800*600*256color mode. but with the issue above, clock frequency becomes a significant limitation:
with 16 colors, both 640*400 at 70hz and 640*480 at 60hz work with 25mhz clock, while 800*600 at 60hz and 56hz work with 40mhz and 36mhz respectively.
but with 256 colors, 640*400 at 70hz and 640*480 at 60hz would require a 50mhz crystal, while 800*600 would require 80 or 72mhz, rendering them unusable on cards without such high-frequency crystals.

now, my idea is attempting to make a round up of all vga cards with such behavior. i initially took only cards with 512kb ram support into account because i expect them to support 800*600*256color mode, or at least 640*480*256colors. many old svga cards have multiple crystals up to 45, 42 or even 40mhz, so they would only run 800*600*256color mode at an extremely low 30~34hz if they ever do, and even 640*480*256color mode at 60hz requires 50mhz, 45mhz would make ~55hz, and 40/42mhz seems unacceptable for me.
but then others reminded me that even cards with 256kb are sufficient for 640*400*256color mode, which is still much better than nothing, and would require a crystal of 50 or 42mhz if they have 4bit video data cycle, which may not be on these low end cards, so they are taken into account aswell.

here are some candidates suspected to have 4bit video data cycle:
trident8800: as already confirmed.
pvga1/wd90c00: suspected from datasheet, but this test result seems to indicate that it runs 640*400*256color mode at 70hz without a 50mhz crystal. however, there seems to be a wd90c60 clockgen whose frequency remains unidentified. also, the result showed that it runs 800*600*16color mode at 56hz using a 28mhz clock rather than the expected 36mhz, wtf?
chips82c450: deduced from datasheet.
chips82c451/441, oak037 and cirrus5401/avga1: these are all 256kb cards and none of their datasheets mentioned any 256color modes higher than 320*200, not even the cirrus5401/avga1 that is believed to be the fastest max256kb card in the world did, wtf? they just claimed to support 16colors up to 800*600 using 40 or 36mhz crystal.
tseng et3000: originally suspected but with no datasheet. after reading manuals of two cards with this chip, i think its probably eliminated.
ati18800: no datasheet.
realtek3102/3103/3105: no datasheet either but suspected from their inherited slow performance. also the 3105 is very common here, but 3102 and 3103 are very rare.