I am using Keil's demo version .... I have written a short kernel

That does not jive. With a 2k limit who in the world would want a kernel.

My guess is that your problem is that you, as many others, think that C is C, it is NOT. Also, the '51 is not a "small PC". For instance the penalty for using int by default when working on a '51 is huge. Another issue is not taking advantage of the special BIT extensions to C. Last, but not least we have the "I am using the large model because I can not be bothered by memory placement". Just two issues that always blow it for the "I think C, not '51" types are function pointers and malloc, buth of which are efficiency killers in a '51 environment.
If you do not study the '51 AND the compilers small differences/limitattions/extensions, you will never be able to write C that is anything but bloated.

You can have your cake (C is C) but you can not eat it too (efficient code).

Erik