Another lookup problem (well-formed vs. valid)

I've been working with data that conforms to a DTD supplied by
my client, Now, they have asked that I use some data from an
additional source which does not conform with their DTD, though
it is parallel (I think) in that the IDs assigned for each entry
are equivalent.

I'm thinking I could use document('....') to pull in the other data,
and I could use <xsl:key> to access the data, but I'm concerned
that the other data doesn't conform to the same DTD as the rest.
I hope (not yet proven) that the other data source is at least well-
formed -- my validating parser hasn't shown me any well-formedness
errors. (By the way, this is off-topic, but does anybody happen to
know how I can just call a non-validating parser to confirm that a
document is well-formed? I have access to xalan/xerces and expat
(nee nsgmls) -- probably some others but not immediately to hand)
Will I be able to just call in the other document without worrying about
its validity?

One more complication: the additional source document is in 27 pieces,
so I was combining them using the internal subset on the DOCTYPE
line, like so:

<!DOCTYPE OPR PUBLIC "-//xxx/DTD xxx//" "Oro-v315.dtd" [
<!ENTITY efa SYSTEM "./ef/xmlout/efa.xml">
...]
>
&efa;
...

An old SGML trick which is still very handy, but here I'm concerned
that I'm no longer in pure well-formedness once I've put that DOCTYPE
line in there. But I don't have access to the internal subset if I don't
have the DOCTYPE line. I suppose I could hardwire the concatenation
together if I have to.