the question is is it practical to generate a cubemap when a entity comes near reflective surface for only a few seconds *nobody willrealize that*

i would take the speed and the time being in the surrounding of a refective surface as a base for rendering a new cube map

if its indoor it shouldn t matter
if its outdoor *e.g.: a reflective water plane i would calculate the frensil effect (spelling?) and see if its visible at all from your view angle

I've heard of all kinds of tricks to making realtime cubemapping faster, i.e. rendering two sides of the cubemap per frame, so the entire cube gets updated every third frame; using a lower bit depth for the cubemap (like 16-bit if you're using a 32-bit display); as Basiror said, only updating it if the camera's (or other objects are) close enough; and rendering the cubemap faces with very close clipping planes. Of course, you'll want to keep the number and size of your cubemaps down as much as possible, no matter what. See if you can use static cubemaps for some things; i.e. things that aren't that reflective and only need a little bit of reflective color.