It depends on a few factors, but assuming you're not using any encounter modifiers like White Flute or Illuminate, and aren't riding the bike, you have an 11% chance of getting an encounter per step in tall grass. And there's a 5% chance that you'll get a Pikachu for any given encounter.
Using binomial distribution, we see that we have a 21.43% chance of getting exactly 2 encounters within 10 steps, with an additional 8.84% change of getting 3 or more encounters, meaning a total of 30.28% chance of >= 2 encounters (though we'll have to treat the >2 scenarios a bit differently).
Pikachu has a 5% encounter rate, so in the exactly-2-encounter scenario, it's a 5% * 5% = 0.25% chance of them both being Pikachu. Accounting for the fact that these odds are only applicable in the 21.43% chance that we get exactly 2 encounters, we'll multiply 21.43% * 0.25% to get 0.0536%, or about 1 in 1867.
However! We still need to take into account the chance that we run into more than 2 Pokémon within our 10 steps. I went and repeated the process above for each of the 3-10 encounter scenarios (using binomial distribution instead of multiplication as necessary) to get a cumulative total of 0.0786% from all of these. Since these are separate scenarios from the exactly 2 encounters we calculated above, we can add the two rather than multiplying them: 0.0536% + 0.0786% = 0.1322%, or about 1 in 757
But that's just the odds of getting two Pikachu within 10 steps, we still need to account for shininess. Technically, there's a whole bunch of additional math we could do to account for running into 3+ Pikachu, and only needing two of them to be shiny, but the odds of that are so unbelievably small and would take so much effort to get that I'm excluding that here; I can guarantee you that it wouldn't noticeably change our answer even if we did find it. So that just leaves us with the simple version:
1/8192 * 1/8192 * our answer above, 1/756.56 for a final answer of 1 in 50,771,882,148
Source for encounter rates per step
Viridian Forest encounter table
