I don't have R, so I can't try this, but this might also work (simplified slightly from your example):

ns <- getNodeSet(xml1, 
  '//DataBank[DataBankName="ClinicalTrials.gov" or 
              DataBankName="ISRCTN"]
   /AccessionNumberList/AccessionNumber')

Would you consider a non-XPath solution?

The XML package has a couple of useful functions; xmlToList() and xmlToDataFrame(). These can convert the XML to native R data structures, which can be easier to work with within R.

Something like this code - which also uses llply from the plyr package to put the accession numbers into a new list.

library(XML)
library(plyr)
url3 <- http://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=21675889&retmode=xml"
xml3 <- xmlTreeParse(url3, useInternal = T)
# convert to list
l <- xmlToList(xml3)

# should really check for existence of DataBankName
# but we'll leave that for now

if(l$PubmedArticle$MedlineCitation$Article$DataBankList$DataBank$DataBankName == "ISRCTN") {
  accn <- llply(l$PubmedArticle$MedlineCitation$Article$DataBankList$DataBank$AccessionNumberList)
}

print(accn)
# $AccessionNumber
# [1] "ISRCTN78147026"

# $AccessionNumber
# [1] "ISRCTN87739946"

It looks unwieldy, but the "$" notation for accessing list elements is helpful, once you see how the XML maps to the list.

There was a typo in my code from the question - extra " and also plural for trial(s).

The last piece of code actually works (repeated here)

(ns <- getNodeSet(xml3, '//DataBank/DataBankName[text() = "ClinicalTrials.gov" or text() = "ISRCTN"]/../AccessionNumberList/AccessionNumber'))