I'm looking for a sparse matrix library that has a C or C++ interface,
preferrably C++ that can perform a QR decomposition using the
Householder transform and subsequently solve the linear system:

Ax = b;
x = inv(A)b;

If A is sparse, I will get quicker solution using the QR transform:

A = QR
x = inv(R)*(Q'*b), where ' represents the transpose

Matrix A is m rows by n columns and m >> n.  It must also be linked
with modules built with MSVC++ 7.0 or 8.0, preferrably 8.0.

I've looked at GNU's scientific library (gsl), Intel's MKL, TNT, MTL
with unsatisfying results.

Basically, I want to mimic the behavior that I see with Matlab.  If
Matrix A is specified as a non-sparse matrix, the computation time is
quite slow for the problem I'm trying to solve.  There is a dramatric
speed-up when A is converted to a sparse matrix.

Have you seen these libraries?
http://flens.sourceforge.net/
http://math.nist.gov/sparselib++/ and http://math.nist.gov/spblas/

I didn't know about FLENS.  While interesting, it doesn't have a
sparse QR decomposition.

I have looked at sparselib++ briefly.  It seems to have a nice class
library for sparse matrices.  The main drawback is the lack of linear
system solvers.

I took a look at the NIST sparse blas library.  It seems to have a few
more functions that the sparse blas library implemented by Intel's MKL
library.  I will look at it further.

One other stipulation that I have is that the object code that I produce 
must be redistributable without license restrictions such as GPL.

Thanks