This totally depends on the primer sets and the taxa in question. Since the sequence lengths that you obtain after 454 are short they usually do not cover most variable regions (usually termed as V1 to V8). Moreover, the reliability of your sequence data should also be considered using the "qual file" that you get along with your sequences. That is, even if your primers cover regions that are differential for the taxa of your interest, if the quality scores are low, then you won't be able to pin down too much hope on the data.

So I reckon, it would be safe to assume till the genus but further down might be dodgy.

So what is the goal of using this method in human microbiome profiling research. There are many publications for that.

Well, it is relatively easier to detect the dominant microbes with non-ngs methods. However, it is now understood that the rare biosphere might play a significant role in systems such as human gut. This is where techniques such as tag-pyrosequencing comes handy.

The simple answer is: it's complicated.

Level of resolution will depend on a lot of variables, but i think the most important are going to be the choice of region and the length of the reads.

Here's something to read that should make the point quite nicely:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2566877/

Now even assuming you go for the 97% OTU identity (which some will argue is a species level resolution) you might find that it's not very easy to transform those otu's into a relevant taxonomy. It's doable, but it isn't easy.

Bottom line, you could get species resolution. Using some fancy sequencing you might even be able to get strain:

https://www.sciencemag.org/content/341/6141/1237439

Good luck!