For example:

$ echo hello | busybox sed 's/l*/@/g'                                           
@he@o                                                                           
$ echo hello | sed 's/l*/@/g'                                                   
@h@e@o@                                                                         

The latter behavior (GNU sed) is the expected, conformant behavior.

It looks to me like sed has an off-by-one error advancing the position to search for the next match after making one replacement.