alternative separate thread implementation of shared variables

 The standard shared variable scheme used by Perl, is based on tie-ing the
 variable to some very special dark magic. This dark magic ensures that shared
 variables, which are copied just as any other variable when a thread is
 started, update values in all of the threads where they exist as soon as the
 value of a shared variable is changed.
 .
 The Thread::Tie module is a proof-of-concept implementation of another approach
 to shared variables. Instead of having shared variables exist in all the
 threads from which they are accessible, shared variable exist as "normal",
 unshared variables in a separate thread. Only a tied object exists in each
 thread from which the shared variable is accessible.
 .
 Through the use of a client-server model, any thread can fetch and/or update
 variables living in that thread. This client-server functionality is hidden
 under the hood of tie(). So you could say that one dark magic (the current
 shared variables implementation) is replaced by another dark magic.
 .
 The Thread::Tie approach has the following advantages:
 .
  * Memory usage - This implementation circumvents the memory leak that
    currently (threads::shared version 0.90) plagues any shared array or shared
    hash access.
  * Tieing shared variables - Because the current implementation uses tie-ing,
    you can not tie a shared variable. The same applies for this implementation
    you might say. However, it is possible to specify a non-standard tie
    implementation for use within the thread. So with this implementation you
    can tie() a shared variable. So you could tie a shared hash to a DBM file à
    la dbmopen() with this module.
 .
 Of course there are disadvantages to this approach:
 .
  * Pure Perl implementation - This module is currently a pure Perl
    implementation. This is ok for a proof of concept, but may need re-
    implementation in pure XS or in Inline::C for production use.
 .
  * Tradeoff between cpu and memory - This implementation currently uses (much)
    more cpu than the standard shared variables implementation. Whether this
    would still be true when re-implemented in XS or Inline::C, remains to be
    seen.