The BEDOPS suite includes a vcf2bed conversion script, if this helps. The bedops tool operates on file streams in linear time and has a low, constant memory footprint, so it will scale to your 10 Gb input file size very nicely (see the Bioinformatics paper and supplementary figures for performance analysis), but you would want to do sorting with the "Big Bed Merge Sort" (bbms) tool, instead of sort-bed, unless you have more than 10 Gb of system memory. (When BEDOPS v2 comes out in a month or so, the sort-bed tool will include the functionality in bbms and be able to do sorts on arbitrarily large BED inputs.)

Please see: http://code.google.com/p/bedops/wiki/vcf2bed for conversion, http://code.google.com/p/bedops/wiki/sortBed for sorting, and http://code.google.com/p/bedops/wiki/bedops for documentation for the bedops tool.

The --element-of operator is probably most useful for reporting overlapping BED elements, while the --merge operator will concatenate overlapping regions. You can combine operators, if this is needed for your analysis, by using standard UNIX piping; BEDOPS apps can usually take in standard input from upstream processing, e.g.

vcf2bed < foo.vcf | bbms - | bedops --element-of - bar.bed > answer.bed