I agree. When working with large XML files, memory can become a concern.
Right off, I must admit that XML::Pastor is not as savy as XML::Twig in this area. As you know, XML::Twig is capable of lazy/selective population of its structures, whereas XML::Pastor will first parse the entire XML into an XML::LibXML DOM tree and then convert the data into Perl native data objects (with hashes, arrays, scalars). On the other hand, the DOM is immediately thrown away so there is no double jeapordy.
In terms of memory usage, XML::Pastor probably compares very much to XML::Simple, because the tree structures are similar except that in the case of XML::Pastor, the references are blessed objects.
Compared to XML::LibXML DOM tree, XML::Pastor probably weighs less in memory as there is quite a bit of book-keeping in the DOM.
One caveat though => currently XML::Pastor is only good for working with DATA style XML (without mixed mark up). This basically means that an element either contains only text or only child elements, not both mixed. So, XML::Pastor is not good for working with a markup language such as XHTML, for example.