Short answer: It will usually screw up.
Long answer: with the "normal" setup (BG uses $0xxx, all sprites use $1xxx), A12 will rise 8 times per scanline (once for each sprite fetch). Only the first of those will actually count the IRQ counter, presumably the others are filtered out and ignored because they're close to each other. However if a gap is stuck between them (such as mixing $0xxx and $1xxx 8x16 sprites) it's possible for the IRQ counter to be clocked multiple (I'd imagine up to 4) times per scanline.
I don't think it's known exactly how close together the rising edges have to happen in order for them to be ignored -- though I'd think it's somewhere between 13 and 24 PPU cycles. I'm pretty sure it can't be less than 13, since one game (don't remember which one) uses $1xxx for BG and $0xxx for sprites, and if you emulate the gap with less than 13 cycles it will count twice per scanline (once at cycle 324, and again at 4 -- instead of just cycle 324 which is intended).
EDIT -- Mario Adventure, a popular hack of Super Mario Bros. 3, mixes $0xxx and $1xxx 8x16 sprites, since at the time, the hack author was unaware of that behavior and emus at the time didn't show the error. I used this game for testing this behavior in my emu. Here is a screenshot of it borking (Nintendulator and NEStopia show similar results):