Cross-graph vs. merged-graph performance

I'm querying across multiple named graphs, and given the shape of the
data it looks like I'll have the choice between -

a) creating a dataset and using queries with multiple GRAPH blocks
b) merging the graphs into one and then querying that as the default graph

Either way I should get the same results. Typically there will be a
single shared node across the graphs, a kind of foreign key (this is
also currently the name of a graph, though that may change).

Is either approach likely to be significantly faster in general, or is
it entirely case-dependent?

I've a feeling this has an obvious answer but seem to have a mental
block on factoring these things out.

Implementationwise I'm using ARQ in-memory (already flipped there from
using MySQL-backed models, there's a possibility of having to flip
back, praise be to common interfaces).

Cheers,
Danny.

-- 

http://dannyayers.com