>if there
> were indeed loans taking place between IE branches, ...,
> any reconstruction attempt we make of that elusive first language
> will be garbled by mysterious alternations we can't account for.
Not necessarily. Sometimes we are able to perceive the basic pattern, and
from that identify the loan word and discount it. Discounting evidence in
order to make a pattern is a dubious practice, but the other way round -
finding the pattern, then treating the aberration as a loan - is not
uncommon. A typical example is the well attested pattern:
medial *bh > b in Latin (ruber etc)
Then the word rufus can be assumed to be a loan from another dialect in
which *bh > f medially. This is secure only because the basic pattern is so
well attested.
A second example is Grimm's/Verner's law. Because this is so well
established, we can assume that words in Germanic langauges that do not show
this are loan words.
But where the patterns are not secure, doubts can remain.
Peter