Thanks for the informative reply.

> ... when you want to work with SIMD or GPU operations, one can (and we will) use a representation of arrays of Nullable objects that places the values and the Boolean flags into two separate arrays.

"When" is pretty much "always" with arrays. Trusting a sufficiently smart compiler and runtime to get rid of the tag checks makes me nervous, though. This optimization needs to be user-predictable and/or -controllable to be truly useful. If only compiler gurus can understand when slow, tag-checking, scalar loops will be replaced with fast SIMD code, something has gone wrong.