The problem is that TCO in the sense we're talking about isn't really an "optimization". It's a language feature that removes the need for explicit looping constructs by allowing you to write them as higher-order functions instead. This only works if the guy writing the higher-order function can prove that the compiler will successfully "optimize away" the tail call, or if he doesn't care whether the program dies with a stack overflow. If the success of TCO is dependent on which conservative approximation the compiler is using for escape and liveness analysis, then people who care about their programs continuing to run will demand explicit looping constructs anyway.

There are three separate inventions in Scheme that work this way. Closures give you objects without an object construct, tail-call optimization gives you loops without looping constructs, and call/cc gives you threads, exceptions, and backtracking without thread, exception, or backtracking constructs.

(I mention Scheme because all three of these were, as far as I can tell, introduced in Scheme, and only later adopted by other functional languages like ML, although to be fair, TCO at least falls out naturally from combinator-graph reduction.)

In a sense, Scheme is sort of like a functional assembly language: there are lots of object systems, looping constructs, threads, and exception systems in Scheme, and they aren't compatible with each other. It's sort of like the situation with linked lists in C, where every library has its own linked-list type.

It's exactly the opposite of assembly language in another way, though. By making object instantiation and population implicit, both the compiler and the maintenance programmer have to do extra work to figure out what's an object and what's not, and what the object's fields are. The same is true of loops, threads, and exceptions. In assembly, instead, you have the problem of things being too explicit, thus losing the signal in the noise.

That's my take, anyway. I haven't spent that much time programming in either Scheme or assembly, although I did write an almost-Scheme compiler targeting assembly called Ur-Scheme.