htmlspecialchars() handles '&' char incorrectly - it doesn't care if it is aready part of entity or not. It results in very "funny" things when this function is being called several times for the same string. For example:

echo htmlspecialchars(htmlspecialchars(htmlspecialchars(htmlspecialchars(htmlspecialchars('text & text')))));

will produce: 
text & text 

Most correct bahaviour will be to check, if it is followed by any valid entity as they're described in HTML specification. However it can be quite hard to do, because there is lots of entities. So another way is also possible (it should be faster but more dirdy): just check if '&' char is started some abstract entity. Here is 2 regular expressions which are implements correct '&' char handling:

1. This is correct way to handle entities:
preg_replace('/\&(?!((#\d{1,5})|(#(x|X)[\dA-Fa-f]{1,4})|[aA]acute|[aA]circ|acute|(ae|AE)lig|
[aA]grave|alefsym|[aA]lpha|amp|an[dg]|[aA]ring|asymp|[aA]tilde|[aA]uml|
bdquo|[bB]eta|brvbar|bull|cap|[cC]cedil|cedil|cent|[cC]hi|circ|clubs|cong|
copy|crarr|cup|curren|[dD]agger|d[aA]rr|deg|[dD]elta|diams|divide|[eE]acute|
[eE]circ|[eE]grave|empty|e[mn]sp|[eE]psilon|equiv|[eE]ta|eth|ETH|[eE]uml|
euro|exist|fnof|forall|frac1[24]|frac34|frasl|[gG]amma|g[et]|h[aA]rr|hearts|
hellip|[iI]acute|[iI]circ|iexcl|[iI]grave|image|infin|int|[iI]ota|iquest|
isin|[iI]uml|[kK]appa|[lL]ambda|lang|laquo|l[aA]rr|lceil|ldquo|le|lfloor|
lowast|loz|lrm|lsa?quo|lt|macr|mdash|micro|middot|minus|[mM]u|nabla|nbsp|
ndash|n[ei]|not(in)?|nsub|[nN]tilde|[nN]u|[oO]acute|[oO]circ|(oe|OE)lig|
[oO]grave|oline|[oO]mega|[oO]micron|oplus|or|ord[fm]|[oO]slash|[oO]tilde|
otimes|[oO]uml|par[at]|permil|perp|[pP]hi|[pP]i|piv|plusmn|pound|[pP]rime|
pro[dp]|[pP]si|quot|radic|rang|raquo|r[aA]rr|rceil|rdquo|real|reg|rfloor|
[rR]ho|rlm|rsaquo|rsquo|sbquo|[sS]caron|sdot|sect|shy|[sS]igma|sigmaf|sim|
spades|sube?|sum|sup[123e]?|szlig|[tT]au|there4|[tT]heta|thetasym|thinsp|
thorn|THORN|tilde|times|trade|[uU]acute|u[aA]rr|[uU]circ|[uU]grave|uml|
upsih|[uU]psilon|[uU]uml|weierp|[xX]i|[yY]acute|yen|[yY]uml|[zZ]eta|zwn?j);)/','&',$str);

2. This is less correct, but still better way to handle them:
preg_replace('/&(?!(([A-Za-z_:][A-Za-z0-9\.\-_:]*)|(#\d+)|(#(x|X)[\dA-Fa-f]+));)/','&',$str);

 Good thing about second regexp is that in a case this way will be implemented by htmlspecialchars() function - it will be possible to use it to handle XML entities aswell.