Write xml-object to disk

I have a big bunch of xml-files, which I need to process. For that matter I want to be able to read the files, and save the resulting list of objects to disk. I tried to save the list with readr::write_rds, but after reading it in again, the object is somewhat modified, and not valid any more. Is there anything I can do to alleviate this problem?

library(readr)
library(xml2)

x <- read_xml("<foo>
              <bar>text <baz id = 'a' /></bar>
              <bar>2</bar>
              <baz id = 'b' />
              </foo>")

# function to save and read object
roundtrip <- function(obj) {
  tf <- tempfile()
  on.exit(unlink(tf))

  write_rds(obj, tf)
  read_rds(tf)
}

list(x)
#> [[1]]
#> {xml_document}
#> <foo>
#> [1] <bar>text <baz id="a"/></bar>
#> [2] <bar>2</bar>
#> [3] <baz id="b"/>
roundtrip(list(x))
#> [[1]]
#> {xml_document}

identical(x, roundtrip(x))
#> [1] FALSE
all.equal(x, roundtrip(x))
#> [1] TRUE
xml_children(roundtrip(x))
#> Error in fun(x$node, ...): external pointer is not valid
as_list(roundtrip(x))
#> Error in fun(x$node, ...): external pointer is not valid

Some context

I have around 500,000 xml-files. To process them I planned on turning them into a list with xml2::as_list and I wrote code to extract what I need. Afterwards I realized, that as_list is very expensive to run. I could either:

1. re-write already carefully debugged code to parse data directly (xml_child, xml_text, ...), or
2. use as_list.

In order to speed up no. 2 I could run it on another machine with more cores, but I would like to pass a single file to that machine, because collecting and copying all files is time-consuming.