It's actually quite FPGA resistant. You need a 700+ buck FPGA to reach performance of a 100 buck CPU in Tenebrix (Already tested by Art, talk to him if you want nitty-gritty details)
Of course, ASIC customized to the task ("lots of fast memory" as you put it) could crunch this baybe fairly well, but for the love of Zod, TBX has to become hellishly expensive and hellishly popular for it to make sense to even try developing such a solution (and implementing it would likely still make no sense, since economies of scale make CPUs very cheap, and it is likely that the number of CPU crunchers you can deploy for the cost of a single "mem-rich" application-specific design would still outperform said application-specific scrypt-cruncher).
So it is quite hostile to at least one common class of "custom" hardware, and another one is plan uneconomical to deploy.
Also bear in mind that CPUs are common household items, even powerful multicoire ones. GPUs suited for mining well are less common. FPGAs are pretty much scientific and special-purpose exotica, and a Tenebrix-optimized ASIC is a theoretic entity, so Tenebrix has a potentially huge miner base, being well suited for the one of the most common types of computational substrate available, and less tendency to have its minerbase concentrated.
And of course there is the fact that CPUs GPUs and FPGAs all retain uses other than "mining some cryptocurrency" while a hypothetical Tenebrix ASIC will be only good for breaking a certain implementation of Scrypt, thus being completely worthless outside Tenebrix network, making a dedicated "Tenebrix cruncher board" even more economically unsound
So to sum up, Tenebrix is GPU-hostile and quite FPGAs hostile, and sufficiently tricky that even powerful ASICs (should such ever be designed for Tenebrix) would likely still be inferior to CPUs (but this time due to costs involved in making those ASICs happen)
All in all, quite custom-hardware hostile, it turns out to be.