When calling get_global_size with dimindx > get_work_dim()-1 it returns arbitrary values. The standard itself says of the get_global size that "For other values of dimindx, get_global_size() returns 1". So I presume this counts as a bug.
As a workaround we currently set the global work dimensions always as 3 for intel platform with unused dimension sizes as 1, however I'd prefer this would be a stopgap measure.
EDIT: It seems that the global id:s are correctly reported. The problem is only with the global sizes. Edited the post to reflect this.