You can use MEGAN (this works with refseq gene IDs). It's for metagenomics mainly, but you might find it useful anyway. Also, take a look at Voronto. This is a new visualization tool which can generate interactive Voronoi maps with cell areas which grow with the number of total annotations (as far as I understood). It works with several built'in ontologies, and you can use your own.

The only solution i can find for you is to use the sigPathway package of bioconductor, most web tools seem to be tailored to eukaryotes.

You can also use InterProScan (v5) to scan your sequences against the InterPro database. InterProScan5 has the options to include both GO terms and Pathways (e.g. from KEGG) associated with your protein by virtue of the InterPro models (aka "protein signatures") that they match.

As an example, InterPro entry IPR011284 incorporates a predictive signature from TIGRFAMs for the enzyme family 3-oxoacyl-(acyl-carrier-protein) reductase. You can see this family has been associated with the GO terms GO:0006633 (fatty acid biosynthetic process) and GO:0055114 (oxidation-reduction process) as well as being implicated in pathways represented in KEGG, MetaCyc and UniPathway (see the "pathways and interactions" sub-page). So, if you search your sequence against InterPro using InterProScan, and this signature matched to it, InterProScan would also show the associations with the GO terms and Pathways in the output.

Hope this helps!

Sarah