optimizing load balance

I've been using pmetis in the trivial (to metis) case of
node weights and no edges. i.e the first few lines in my
weight file are:
1155 0 10
with a max value of 519.692 and a min value of 7.15386
(the numbers are my proxy measure for neuron complexity
(499 cells broken into 1155 pieces) and I need to
place the pieces onto 1024 cpus in a balanced fashion).

Anyway, pmetis yields
METIS 4.0.1 Copyright 1998, Regents of the University of Minnesota

Graph Information ---------------------------------------------------
Name: balbb520.wt.dat, #Vertices: 1155, #Edges: 0, #Parts: 1024

Recursive Partitioning... -------------------------------------------
1024-way Edge-Cut: 0, Balance: 1.55

Timing Information --------------------------------------------------
I/O: 0.000
Partitioning: 0.010 (PMETIS time)
Total: 0.010

whereas a naive greedy method (sort the weights
and fill each cpu in turn using the maximum unused
weights that do not sum greater than some threshold)
gives a balance of 1.17. So I am wondering if I am
misusing METIS and if there is a way to cast the problem
to reliably get nearly optimum load balance.

RE: Metis's heuristic algorithms

Metis's heuristic algorithms have been optimized for partitioning large unstructured graphs and they almost always fail to produce good solutions for relatively simple problems.

So to answer your questions, Metis cannot solve such problems in a nearly optimum way and it gets misused in the sense that it was not designed for that.