After doing some quick back-of-the-napkin calculations, I might actually have changed my mind a little.

If we assume that MN and minter are paid the same amount out of each block (either 50/50 or 45/45/10 (mn/minter/devfund)), what is important is to find a level where there's slightly more coins staking than locked in masternodes. As long as this is fulfilled it would be profitable to run a MN.
Current net-stake-weight is ~28 million, so the tipping point where staking is more profitable than running a MN is: 140MN @ 100k collateral, or 1,400MN @ 10k

In other words, the smaller the collateral the more MN can be profitably supported by the network, as long as the amount of coins staking is larger than what is locked in masternodes.
So the golden number is whatever ensures that enough coins keep staking.

Anyways, enough of me thinking out loud.

PS. Another thought. If we want to make it possible for "everyone" to afford a MN, then restrict masternodes to run a specific port, like ARC. Considering the low cost, and the profit per MN, means that it's not profitable to rent a VPS to run a MN on, and thus limiting the number of total masternodes.

PPS and now I stop thinking for real.