Here's timing produced by the five statements in the following function, to give you a sense of relative speed on my computer:

   user  system elapsed 
 12.227   0.871  13.100 
   user  system elapsed 
 41.688   3.735  45.423 
   user  system elapsed 
636.549  29.291 665.827 
   user  system elapsed 
  5.991   0.060   6.052 
   user  system elapsed 
164.827  62.318 227.457 

writeIntronFlankingSeqs<-function(genomeName='BSgenome.Hsapiens.UCSC.hg38'
          ,TxDbName='TxDb.Hsapiens.UCSC.hg38.knownGene'
          ,flankSize=25
          ,con=paste(sep='.','sjdb',flankSize,TxDbName,'fa')
           ) {
    ## PURPOSE: Write a fasta file holding <flankSize> bases flanking
    ## each unique intron known to <TxDbName>, as extracted from the
    ## BSgenome named <genomeName>.
    library(rtracklayer)
    library(GenomicFeatures)
    library(BSgenome)
    genome<-getBSgenome(genomeName)
    require(TxDbName,char=T)
    TxDb<-eval(parse(text=TxDbName))

    print(system.timeintron.gr<-unique(unlist(intronsByTranscript(TxDb)))))

    print(system.time(
      flanking.grl<-
        ## Kudos: C: Element-wise setdiff when psetdiff fails for GenomicRanges
        setdiff(as(trimintron.gr+flankSize),'GRangesList')
               ,asintron.gr,'GRangesList'))))

    stopifnot(all(2==lengths(flanking.grl)))

    print(system.time(
      flanking.grl<-
        ## Required for correct 5-3' order in subsequent call to
        ## extractTranscriptSeqs.  Kudos:
        ## [[Reorder exons on negative strand in GRanges list][Reorder exons on
        ## negative strand in GRanges list]] - along with appeal for
        ## faster approach (THIS IS SLOW).
        revElements(flanking.grl,any(strand(flanking.grl)=='-')) 
    ))

    print(system.time(
      names(flanking.grl)<-
        ## Creates names of the form:
        ## chr:start1-end1:strand^chr:start2-end2:strand
        do.call(function(...) paste(sep='^',...)
               ,as.data.frame(matrix(as.character(unlist(flanking.grl))
                                    ,ncol=2
                                    ,byrow=T)))
    ))

    print(system.time(
      writeXStringSet(extractTranscriptSeqs(genome,flanking.grl)
                     ,con)
    ))

  }

Thanks for taking a squiz Michael.