So I've found a bit more info.
In order for any of this to work, bit 4 of address 0xc8e2 must be set. This would correspond to whether you've visited that tile, but it's overflowing into some other data. I don't know what this corresponds to, but a 100% completed game has this set. (edit: or, maybe just walk there? :P)
The primary factor which determines the effect, is the variable at address ff8d. This corresponds to the x-position of a sprite. If there's no time portals or anything on the screen, it corresponds to link's x-position.
It goes through a loop with this variable, and each iteration, it decrements the value. So, values 84-89 (for instance) have the same effect because they all eventually execute the loop with [ff8d] == 83. This particular value has a glitchy effect which ends the loop somehow.
Here are the values I've tested... some effects seem to vary between games:
06-12 = nothing (sometimes changes music tempo ???)
13 = crash
14-22 = nothing (?)
23-25 = shooting gallery "strike" sound effect (sometimes crash?)
26-2f = crash (invalid opcode)
30-3c = crash (because of 2f)
3d-5e = crash (because of 3c)
5f-77 = crash (because of 5e)
78-79 = graphical map glitch
7a = graphical map glitch + get rupees (???) (second time crashes)
7b-81 = crash (because of 7a)
82-83 = crash (different cause)
84-89 = graphical map glitch (1st time), use A button item (subsequent times)
8a-92 = graphical, audio glitches (1st time), crash (subsequent times)
93-9a = crash (because of 92)
That's the range of link's possible x-positions. It might be possible to get higher values (I don't think lower would matter).
There's not much particularly useful so far, but perhaps some of those crashes could be studied further, who knows.