You'd think that the 9918 would be a bottle neck between the CPU and the VRAM, but the Apple II can write to the 9918 VRAM as fast as its own internal RAM. Faster, actually, since the destination address auto-increments. That was a surprise.
Hmmm, is the bitmap linear by line, or C64 style?
Auto increment + C64 style would rock pretty hard. It's still good per line.
Both have their merits.
A single screen is divided into 3 separate tile maps, each 256 bytes long.
It also supports 32 spirtes but only 4 sprites per scanline, which is its biggest limitation.
The Sega Master System used an upgraded version of the 9918, adding 64 colors, smooth scrolling, and 8 sprites per scanline, all within the same 16 kB of VRAM.
Yeah, auto-increment can rock! Blasting characters to the memory would be fast, and only require a precise index for the source data. Perfect for the 6502. Other index can be done every so often, depending on what is being drawn.
Overall, given memory expansion, the goofy screen didn't end up being too big of a deal. Most programmers made Y axis lookup tables and called it a day. The fact that the artifact color mapping repeated every word, two bytes, did have an impact in that it was generally faster to maintain two pre-shifted copies of software sprites though.