No idea about how this compares speed-wise, but python is a bit more flexible and easier to use when filtering flags, I find. The pysam packages uses a wrapper around samtools to read in and write out BAM files.

I wrote a script once for checking all kinds of flags, tried to strip it down to what you might be interested in (haven't tested it though)

#!/usr/bin/env python

import sys
import argparse
import getopt
import pysam
import os

def get_args():
    parser=argparse.ArgumentParser(description='filter BAM files')
    parser.add_argument('--BAMfile', '-in', type=str, required=True, help="BAM file")
    parser.add_argument('--outfile', '-out', type=str, required=True, help = 'Name for output file')

    args=parser.parse_args()
    return args

def remove_multiAlignments(ReadTagsEntry, KeepInfo):
    '''Checks for NH and XS tag to remove reads that aligned more than once'''
    
    keepRead = KeepInfo
    
    if 'XS' in ReadTagsEntry:
        keepRead=False
    elif 'NH' in ReadTagsEntry and ReadTagsEntry[1] > 1:
        keepRead=False
    
    return(keepRead)

def main():
    args = get_args()
    
    infile = pysam.Samfile(args.BAMfile, "rb")
    out = pysam.Samfile(args.outfile , "wb", template = infile )
    
    keep = True

    for DNAread in infile.fetch():
        intags = DNAread.tags
         
         for entry in intags:
            keep = remove_multiAlignments(entry, keep)

        if keep:
            out.write(DNAread)

        keep = True

    out.close()

if __name__ == '__main__':
main()

The filter function from Bamtools (https://github.com/pezmaster31/bamtools) may be a better choice here. Since the bam file is no decompressed, it should be faster than grep.