We should make this more available to the programmer! It should be possible for a programmer to highlight a section of code in an IDE and choose "Apply Trace," whereupon the IDE will apply static type annotations based from saved runtime trace information. The IDE should be also able to present the same information as profiling data to let the programmer quickly home in on the 15% or so of the app which is most performance critical. Using techniques like this should let us get C-like speeds from many dynamic languages.
(Admittedly, this sort of thing would also be highly dangerous. One couldn't apply such semi-automated annotations in ignorance. This could also break otherwise correct programs.)
In other words, a good programmer should be able to just let the VM know ahead of time what's up in critical sections of stable production code. Even better, we should be able to combine datasets from many different runs of many VM instances! This sort of technique would allow us very high degrees of confidence for things like web apps in server farms, where getting 10's of thousands of runtime tracing datasets would be easy to do.
This is where a dynamic language with optional type annotations would really shine.
Snippet from [0], which is unfortunately temporarily down (I got to it through Google Cache[1]):
cpdef double integrate_f(double a, double b, int N):
cdef double dx, s
cdef int i
dx = (b-a)/N
s = 0
for i in range(N):
s += f(a+i*dx)
return s * dx
[0]: http://www.behnel.de/cython200910/talk.html[1]: http://webcache.googleusercontent.com/search?q=cache:7NvoUHh...
Also: Common Lisp, Scheme, Erlang, SML, OCaml, Haskell, Lua, Prolog, and Java, and no doubt numerous others.
For all of the above, many (if not most) implementations compile natively.