Problems I see in your code:

• Your algorithm is fragile; it depends on the number of $user fields (scattered in %onprop) to match the same number of $user fields (clustered in @dataarray). It took me quite a bit of reading to understand this dependency, so I would say it will be 'non-obvious' to your maintenance programmers. If the lines in PAGE should change their sort order in the future (un-clustering the $users in @dataarray), or code is added to pre-filter the $users in only one of the two data structures, then your code will fail badly and *invisibly*, shifting away data with no warning that anything is amiss.
  The grouping problem is usually solved with a hash of arrays (HoA); see my code below.
• Your algorithm is wasteful; you must loop through %onprop, with hash lookups inside the loop, (keys(%onprop) * number_of_distinct_users) times. Also, your unless (! $item) could be moved outside the foreach my $scalar loop.
• Two problems with my $a = shift(@dataarray);
  1. The use of $a or $b outside of sort blocks is frowned-upon, due to their semi-magic.
  2. Since you are throwing the data away, it is better written without the receiving variable (void context), as in shift @dataarray;
• As others (++ to all) have pointed out, huge lists of scalar vars is non-optimal. @field_names and hash slices allow you to get rid of mystery numbers like 4 in @{ $onprop{ $scalar } }[4].
• When you said while (<@dataarray>), I shuddered. It would probably never fail in this particular program, but you should use while (@dataarray) instead. This should illustrated the difference:

  $ perl -MO=Deparse -e 'while (@d) {}'
  while (@d) {
      ();
  }
  
  $ perl -MO=Deparse -e 'while (<@d>) {}'
  use File::Glob ();
  while (defined($_ = glob(join($", @d)))) {
      ();
  }
  [download]

Here is my re-write of your core code, using HoA grouping and hash slices. Working, tested code:

#!/usr/bin/perl -W
use strict;
use warnings 'all';
use Data::Dumper; $Data::Dumper::Useqq = 1; $| = 1;

my @field_names = qw(
  PROP
  COLOUR
  TXTCOL
  URL360
  USER
  YOUR_NAME
  ADDRESS
  TOWN
  ZIP_CODE
  COUNTRY
  EMAIL
  TELEPHONE_NO
  TELEPHONE_NO2
  THEME
  WEB_ADDRESS
  PPEMAIL
  JUNK1
  JUNK2
  JUNK3
);
# Simplified @field_names for this example:
@field_names = qw(
  PROP
  COLOUR
  USER
  YOUR_NAME
);

# 'Prop' is a unique key in file PAGE and in hash %onprop.
my %onprop; # HoH

# 'user' is not unique, so each hash entry will be an array of 'Prop's
+.
my %user_props; # HoA

while (<DATA>) {
  chomp;
  next unless /\S/;
  my @fields = split "\t";
  warn unless @fields == @field_names;
  my %h;
  @h{@field_names} = @fields;
  my $prop = $h{PROP} or warn;
  my $user = $h{USER} or warn;

  # Create array of users and hash of data
  warn "Overwriting '$prop'" if exists $onprop{$prop};
  $onprop{$prop} = \%h;

  push @{ $user_props{$user} }, $prop;
}
print Data::Dumper->Dump(
  [     \%onprop, \%user_props   ],
  [ qw(  *onprop   *user_props ) ],
);

print join("\t", @field_names), "\n\n";
foreach my $user ( sort keys %user_props ) {
  my @prop_list = @{ $user_props{$user} };
  my $number_of_props = @prop_list;
  print "Grouped data ($number_of_props lines) for user '$user'\n";
  foreach my $prop (@prop_list) {
    my %h = %{ $onprop{$prop} };
    my @fields = @h{@field_names};
    print join("\t", @fields), "\n";
  }
  print "\n";
}
__DATA__
foo    red    jonnyfolk    Saint J
bar    white    Util    Me
baz    blue    jonnyfolk    Saint J
qux    black    Util    Me2
[download]