I forgot to clearly state an assumption I made, which was that I wasn't working the numbers based on averages but on near certain success. At that point, assuming that over 90% is "near certain success," you need 4 dice to get at least a 90% chance to roll a single success. if you raise the number of successes required from 1 to 2, then you need 8 dice to have a better than 90% chance to roll 2 successes. That's where my +4 dice number comes from; it's not an average, it's a high-certainty value. I think I muddied things up by then talking about 3 and 4 successes, which I really meant "giving the player the equivalent bonus of 1 success is tantamount to giving them +4 dice" (but if you're talking about averages then yes, it's somewhere slightly south of 2 dice worth of bonus).

Here's my Anydice program to support the numbers.

I believe the conclusion is still valid: getting a bonus of 1 success is worth more than getting an extra die to roll, at least in this example, rolling 50/50 dice.