> but AFAIK there is no rigorous proof yet that the selfish mining equilibrium isn't a real threat.
It's very easy to detect and respond to. Point your miners at the large pools, and check if they are building off of non-public blocks. Respond accordingly.
How do you know they're building off of non-public blocks if they're doing so in secret collusion?
For what it's worth, Felten et al have a somewhat convincing (if not yet rigorous) argument that members of a so-called selfish mining pool have more incentive to defect than to continue cheating [0]. But, if a single entity controlled all of the 33%, defection is not a risk, and this entity could still claim an outsized share of the mining reward according to Eyal-Sirer.
Who are the miners? Mining pools? Mining pools don't do the actual mining, they contract that out to their users. Anyone can run mining software pointed at any of the major pools and see what blocks they are building off of. There's no way for a mining pool to "build a blockchain in private."
If not mining pools, then perhaps the concern is large hosted operators? There's only one of these that qualify, although there is insufficient data about their exact size (CEX.io, part of GHash.io, but only an unknown percentage of their hashpower is hosted).
Mining is sufficiently decentralized so as to make large-scale selfing mining detectable, and the economics are such that small-scale selfish mining is unprofitable. I'm not claiming this will always be the case - there is sufficient cause to be concerned for the future as we are trending towards more centralization and hosted mining, not less - but it is certainly not a real concern now.
It's very easy to detect and respond to. Point your miners at the large pools, and check if they are building off of non-public blocks. Respond accordingly.