So I've investigated the glitch brought up here a bit more:
http://tasvideos.org/forum/viewtopic.php?p=400831#400831
Turns out the reason for the memory corruption is because power bombs affect the clipdata and appearance of certain blocks. For example, when a power bomb hits a bomb block, it changes its clipdata to 0 (air) and its BG1 value to 0 (clear). When the explosion touches a particular block, it uses the following calculation to find the location of its clipdata value in memory:
0x2026000 + ((Ypos / 64) * RoomWidth + (Xpos / 64)) * 2
where 0x2026000 is the start of a room's decompressed clipdata, (Ypos / 64) is the Y position in blocks, and RoomWidth is the room's width in blocks.
Now, when you go out of bounds, your X and Y position can go below 0 and wrap around to 0xFFFF, which is 0x3FF in blocks. This causes the game to read memory addresses way past that of the room's clipdata.
So, is there anything of interest located in memory past the clipdata? Slot A save data (SRAM) starts at 0x2038200. Is this address reachable? Doing some quick math, you would need to go out of bounds in a room that is at least 0x24 blocks wide.
Unfortunately, this is where things get less exciting. Power bombs only affect blocks with a clipdata value between 0x0050-0x0077, which consists of various breakable and exposable blocks. This means that in order to corrupt memory, the first (even) byte must be between 0x50-0x77, and the second (odd) byte must be 0. Furthermore, this only changes the first byte to 0. In terms of breaking the game, this glitch is useless. However, you could do something silly, like resetting the in-game time to 0 once you reach 0x50 (80) hours. Of course, corrupting SRAM is only possible if you can get out of bounds in a wide enough room. Currently, it's only possible to get out of bounds in Yakuza's room, which isn't wide enough.
Also, it's not possible to use the new BG1 value of a block in order to corrupt SRAM. When the address of the block is calculated, only one byte is used for the X and Y block position, not two.