xsi:schemaLocation should contain a list of URIs, semantically grouped two by two. Each URI appearing at an odd position specifies a namespace, and the URI appearing at the next even position specifies the location hint of the schema to use for this namespace. Location hints may be local or remote.
This is an example with three namespaces:
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<ord-archive:XXX
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xmlns:ord-archive="http://www.example.com/namespace"
xmlns:prefix="http://www.example.com/another-namespace"
xmlns:other-prefix="http://www.example.com/and-another-namespace"
xsi:schemaLocation="
http://www.example.com/namespace
http://somewhere.example.com/something.xsd
http://www.example.com/another-namespace
some-relative-path-to-a-directory/some-local-file.xsd
http://www.example.com/and-another-namespace
file:///some-absolute-path-to-a-directory/some-local-file.xsd">
<!-- the document itself, which contains elements in the three
namespaces, referenced with the bound prefixes. -->
</ord-archive:XXX>
The XML Schema specification leaves some degree of freedom to validation engines, so in theory they may or may not follow the hints, they may use cached versions for some namespaces, etc. However, in practice, most validation engines will use your hints or provide an implementation-specific way of controlling their behavior.
The example given under the link in the question uses xsi:noNamespaceSchemaLocation, which contains a location hint for a schema with no target namespace.
