I also uploaded a jpeg of the ROM array rotated (CCW) to match the other dice, with a square overlaid on each 1 bit from my transcription: www.seanriddle.com/ti57rombits.jpg I've checked it several times, but I wouldn't be surprised if there were a couple of stray bits.
It looks like the physical layout is pretty simple; the 13 columns of 16 bits per row are the 13 bits of each word (MSb on the left) for each page, and the rows are in order. The only complication is that the page order is reversed every bit- bit 12 has them in order 0-F, but bit 11 has them reversed F-0. I uploaded this ordering as www.seanriddle.com/ti57.bin
I compared it to the ROM dump in patent 4125901, and it is very similar; Hrastprogrammer told me they were different, so that was expected, but I didn't know how similar they would be. It turns out that most of the calls and branches are 2-3 bytes different, but the other opcodes are generally the same.
I'll get a TI57 eventually and try to dump the ROM electronically to compare. I picked up a TI55, which uses the same chip, and I'll dump it, too.