See CR 6971190 where the similar problem is described.
With the same schema as in CR 6971190 the following xml document will be accepted since JDK 7 b126:

<?xml version="1.0"?>
<doc xmlns:xsi='http://www.w3.org/2001/XMLSchema-instance&#39;
    xsi:noNamespaceSchemaLocation='reS51.xsd' >

<!--
base='string', pattern='\d', value='#x0BE6;', type='invalid', RULE='37'
-->

      <elem att='&#x0BE6;'/>

 </doc>

The "x0BE6;" is "TAMIL DIGIT ZERO" according to Unicode 6 ( http://www.unicode.org/Public/6.0.0/ucd/UnicodeData.txt) which was integrated into JDK 7 since b 121(See CR 6959267).

The issue also exists for the following symbols:
x0BF0 - TAMIL NUMBER TEN
x0F2A - TIBETAN DIGIT HALF ONE
x1372 - ETHIOPIC NUMBER TEN

Regression since JDK 7 b126. Looks like caused by jaxp 1.4.5 was integrated (See CR 7007257).