Here are the results of simulating 100,000 "draw X and choose 2" drafts, categorizing the playability of the best pair(s) in the selected packs using the playability matrix described in the overview:

"Draw three" is disqualified immediately, because in roughly 1 in 100 drafts you'll end up with zero "playable" decks in your pool of packs. That doesn't seem like a lot, but having any games that are purely guaranteed to be miserable just feels bad. Beyond that, you'll only end up with a medium-playable deck option 66% of the time and a strongly playable option 42% of the time.

It looks like "Draw Four", "Draw Five", and "Draw Six" are all reasonable. With each you have an over 90% chance of dealing at least a medium-playable pairing. Between the three, I prefer "Draw Five". You have a 98.8% chance of getting at least one medium-playable pairing and a 87% chance of at least one strongly playable pairing. On top of that, between two players you'll only "see" half of the total pool, increasing the variety of decks you're likely to see across multiple plays compared to a "Draw Six" strategy that would allow repeat decks to be picked more regularly. Finally, "Draw Five" allows you to simply deal five packs to each of four players if you'd like to use the cube for tournament-style play.

There's one more draft strategy I like and would recommend for people who are very comfortable with the set and its synergies: "Draw Three Twice"—where you pick one of three packs, then deal out another three, and pick your second pack from among those three. From a purely random perspective, the probabilities here are slightly worse than the "Draw Three" drafting strategy. Where it differs is when you account for player skill in making that first pick, knowing how likely you are to see synergistic pairs between that pack and the next three. There's higher risk/reward factor because picking a two-colour pack gives a slight chance of having only unplayable pairings available in the second three, while picking a mono-colour pack is "safer" but gives fewer chances at a strongly playable deck.

The script I used to generate this analysis is publicly available here.