2-3W/port may be not an issue for server blades as not many of them are used. But it could be a big issue for ToR switch which may have up to 60 of them in 1U. I guess you have to use the same type of connectors on both ends of a cable especially in the case of an electrical or an optical cable. Considering a 10Gbase-SR consumes much less power than a 10Gbase-T.
I agree that with 2-3W/port power consumption as not a big deal (the first generation PHY's were in the 10W/port range). In comparison the FCoE ports are in the 1-1.5W/port range and also have advantages of supporting two different fabrics.
The power consumption in the 5W/port range is a concern for server blades but even those can be adequately designed for thermal margins.
that last comment about PHY power needs some justification. why would 2-3W/port be considered a big deal? we're talking about a couple ports per server, so that's about 1% of the power budget.
price (incl switching) is the big problem. especially since for HPC, neither latency nor bandwidth are competitive with IB.
The ASPs have to go further south of $200 per 10GBase-T ports for the base-T market to proliferate.
The power issue has been a well-known concern for years now, I hope the smaller node and less power hungry designs can alleviate it.
What the article did not mention is the fact that many data centers are already wired for 10G base-T BUT have not yet fully utilized the fabric. It would be nice to know what percentage of 10G base-T fabric is still using the 1G appliances.