Hi,
as far as I know, the main difference is that a local aligner just tries to align a portion of the query sequence and tries to extend the alignment; on the contrary, a global aligner tries to align the whole query sequence. Still, if there are too many differences at the 5' or at the 3' of the primary alignment, those regions may be soft-clipped, namely unaligned (decreasing the query coverage). In practice, I don't know if this implies a huge difference is expected in the results, probably not.
Regarding the results, I think that the differences in the processing time could be explained by the default value of --p-maxrejects set to 'all'. That value may be set to a lower value (e.g. 100), in order to limit the number of sequences in the database, presorted based on k-mers in common with the query that are aligned to the query, as suggested in this post .
For example, you may try out these parameters:
--p-maxaccepts 100 \
--p-maxrejects 100 \
--p-maxhits $MAX_ACCEPTS \
Also, setting such a low --perc-identity coupled with MAX_ACCEPTS=10 probably results in top hits belonging to very diverse taxa, causing a high-level consensus taxonomic assignment. Since you are setting such a low identity value, you are aligning Nanopore reads, right?